A Cost-Efficient Multi-Angle Fusion Deep Learning for Ultrasound Localization Microscopy.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Ultrasound localization microscopy (ULM) provides super-resolution imaging of microvasculature.
• Current clutter filtering methods, like SVD, are computationally demanding and hinder real-time ULM.

Purpose of the Study:

• To introduce AF-UNet, a deep learning framework to accelerate clutter filtering in ULM.
• To enable real-time, clinically applicable microvascular imaging.

Main Methods:

• Developed AF-UNet, a lightweight, multi-angle deep learning model for clutter filtering.
• Processed rotated 3D in-phase/quadrature data to suppress tissue signals and reconstruct microvascular volumes.
• Analyzed optimal angular acquisition settings for enhanced fusion performance.

Main Results:

• AF-UNet achieved over 20-fold computational speedup compared to SVD filtering.
• Demonstrated robust performance and generalization across brain, eye, and kidney imaging.
• Preserved microvascular details with high image fidelity.

Conclusions:

• AF-UNet offers a computationally efficient solution for clutter filtering in ULM.
• The framework facilitates real-time microvascular imaging with preserved image quality.
• Optimized angular settings improve fusion performance for diverse anatomical regions.